We are excited to share preliminary results from our narrative factor alpha benchmark.

Applying a naive, systematic implementation of Hivemind's narrative factor within a S&P 500 proxy universe, our backtest achieves:

• a 1.0x IR in an industry-neutral, top minus bottom decile portfolio, and
• a 0.9x IR in a subindustry-neutral, top minus bottom decile portfolio.

Preliminary Backtest Results

Industry-Neutral

# Initial date                                       2020-01-02
# Final date                                         2026-03-18

# Annualized excess return (%)                            6.49%
# Annualized excess risk (%)                              6.42%
# Information ratio (x)                                   1.01x
# Max drawdown (%)                                        9.89%
# Max daily turnover constraint (% AUM)                   3.00%

Subindustry-Neutral

# Initial date                                       2020-01-02
# Final date                                         2026-03-18

# Annualized excess return (%)                            4.48%
# Annualized excess risk (%)                              5.03%
# Information ratio (x)                                   0.89x
# Max drawdown (%)                                        6.74%
# Max daily turnover constraint (% AUM)                   3.00%

The above backtests, built and run using our open-source solution ForecastOS PM, exclude transaction costs and short holding costs.

Turnover is ~3.5x (two-way) per year, or roughly 1.75x per (long and short) side.

Hivemind's Narrative Factor: What is Being Measured

Our narrative factor attempts to capture performance drivers similar to the momentum factor, but instead of chasing a symptom (i.e. previous returns which may indicate the presence of a positive tailwind), it chases the root cause - growing market-relevant narratives driving changes in expectations and prices.

From Narratives to Numbers: Measuring Belief with Hivemind

Hivemind is our trend identification and company exposure engine. It measures discussion to discover market relevant trends and scores trend impact, with direction and magnitude, for every company in your universe. All processes and outputs are customizable and point-in-time. Built for institutional…

The ForecastOS BlogCharlie Reese

The Hivemind narrative factor blends discussion-weighted Hivemind exposures related to point-in-time market relevant narratives within 95% of their trailing 6-month discussion highs.

The result is a new factor that is broadly applicable and that works on a medium to long-term timeframe.

You will notice a couple of short horizontal lines in the above excess return evolution. Those correspond to brief periods when there were no market-relevant, investable narratives within 95% of their 6-month discussion highs. Said another way: the Hivemind narrative factor is a weighted composite of transient narratives and associated exposures; occasionally, no market relevant narratives near trailing highs exist.

Hivemind's Narrative Factor: Why it Matters

While preliminary, the results are unusually strong given the liquidity and efficient pricing of the S&P 500 universe. Further, all else equal, increasing breadth should increase IR.

While more research and proof of value is due, we're excited by what we've seen to date!

Accessing Hivemind's Narrative Factor and Backtest

ForecastOS Hivemind clients have access to the code and methodology used and discussed above, including the portfolio construction framework, narrative factor construction / eligibility rules, and performance measurement approach used in these preliminary results.

Clients can inspect exactly how the benchmark is built, reproduce the process end to end, and adapt the implementation to their own research, universes, and deployment constraints. Our goal is not just to show the results, but to make the benchmark transparent, testable, and directly usable as a foundation for live investment workflows.

If the above is of interest to you, email us at hi@forecastos.com to schedule a call. We'd love to show you what we've built and how you can integrate it into your research and investment process.

Note: associated white paper to come.

Market-moving themes like GenAI, geopolitical conflict, inflation, and tariffs are poorly captured by risk and alpha models.

If you ran a large book through 2025's volatility, you felt it. You'd be forgiven for forgetting that the same forces that have always driven markets and valuations remain intact: expectations of future cash flows.

Anyone who has built a discounted cash flow (DCF) model would know that intimately. Assuming a typical discount rate, the majority of implied company value is based on assumptions and narratives about what the world will look like in 4-20 years.

And those narratives/expectations, about what the world will look like in 4-20 years, have been changing faster than ever before. Unfortunately, systematically quantifying point-in-time changes in market expectations, and the direction and magnitude of companies impacted, is impossible… or is it?

More on that below RE: ForecastOS Hivemind. It's certainly been impossible until recently.

Given the challenge of measuring changes in aggregate long-term expectations (which comprise the majority of implied company value), investors have focused on shorter-term sources of return, including:

• Event-driven forecasting; fundamental, macro, and other outcomes over the next 0-2 years
• Market abnormalities; exploited on aggregate via short-horizon strategies
• Structural and thematic analysis; trade imbalance, regime change, etc.
• Fundamental analysis; better understanding of business models or near-term headwinds/tailwinds and inflection points

But none of the above measure shifts in longer term expectations/narratives, which have driven recent volatility and returns. Most of what drives prices remains poorly modelled by institutional investors, or not modelled at all.

Fortunately, thanks to recent advances in data availability, embedding models, and GenAI, you can now measure shifts in market narratives and reason through which companies are positively and negatively impacted using forward-looking perceived causality, not backward-looking correlation.

We do that. We call our solution Hivemind.

Hivemind is our trend and exposure engine. It takes messy, unstructured information (filings, podcasts, any text/financial data - even your own proprietary inputs) and turns it into clean, point-in-time exposures/factors. And because Hivemind is built to track narratives, it also continuously surfaces and ranks market-relevant trends, both daily...

and grouped across time...

... so you can see what’s emerging, what’s fading, and what’s quietly becoming a persistent factor.

Describe a concept once and Hivemind will score every company in your universe with direction and magnitude, at each point-in-time - so you can separate helped by vs hurt by, how much, and when.

Hivemind doesn't use a static taxonomy. You can tune the “recipe,” swap datasets in/out, and dial the scoring logic up or down (the “knobs”), then export it in the result schema you want - via UI or API - in minutes.

We built Hivemind specifically for institutional workflows. It’s already been used in the wild for neutralizing portfolio themes (e.g. GenAI, inflation, tariffs) against a benchmark, applying event-driven thematic exposures, and more.

Venezuela Hivemind Trend Generates 5% Excess Return Since Inception

Military escalation with Venezuela / Maduro has been the #4 market relevant Hivemind trend since Sep 25, 2025; surpassed only by GenAI, Trump, and Middle Eastern conflict. Given Maduro’s capture on Jan 3, 2026, we wanted to share how our associated Hivemind exposure performed in excess of market an…

The ForecastOS BlogCharlie Reese

But don't just take our word for it. See what Jonathan Briggs, CIO of hedge fund Tc43 and ex-Head of the Alpha Generation Lab at CPPIB, has to say:

"Traditional risk models remain necessary but are no longer sufficient. Market narratives and investor behavior now generate disruptive price dynamics that conventional frameworks weren't designed to capture. Tracking sell-side thematic baskets - often just a handful of names whose relationships are statistically unstable and decay quickly - provides little value for systematic managers trading thousands of instruments across multiple asset classes. Hivemind bridges this gap. By transforming complex unstructured and structured data into dynamic, point-in-time thematic exposures - calibrated with investor insight and updated systematically - it gives quantitative investors the tools to manage risks that legacy models simply can't see. This is the first real step toward next-generation risk control."

To demonstrate one of many potential applications, check out the below Hivemind S&P 500 macro overlay for a long-only portfolio with:

• 2.5% excess return annualized
• 1.0x information ratio
• 3.4x one-way turnover

Needless to say, we're excited! To those we've spoken or worked with thus far: thank you. We're a small team, but we'll continue to do our best to get Hivemind into more of your hands over the coming months.

To learn more about Hivemind, email us at hi@forecastos.com to schedule a call. We'd love to show you what we've built and how you can integrate it into your investment process.

More on Hivemind:

ForecastOS Hivemind: Understand Your Exposures

If you ran a large book through 2025’s volatility, you probably spent more time on the phone than in the market. “Are you down? We’re off 3%.” Most of those calls were really a fishing expedition for: “what exposure did I miss?” and “does everyone have it, or

The ForecastOS BlogCharlie Reese

Hivemind Factors Explain Incremental Out-of-Sample Variance

Table of Contents 1. Why 2025 Has Been Challenging For Quantitative PMs 2. Creating Macro and Subindustry Hivemind Factors 3. How We Measured Out-of-Sample Explained Variance (R²) 4. Results: Hivemind Factors Explain Incremental Out-of-Sample Variance 5. Email Us: hi@forecastos.com 💡Hivemind…

The ForecastOS BlogCharlie Reese

ForecastOS Hivemind: Measuring Emergent Sources of Risk & Return Since 2020

Table of Contents: 1. ForecastOS Hivemind 2. Emergent Themes; Top Sources of Risk & Return Since 2020 3. Closing Thoughts Now, more than ever, the stock market is driven by emergent themes: * The rise of generative AI * Rapidly evolving inflation environments * Wars, both trade and actual…

The ForecastOS BlogCallum Baker

Given Maduro's capture on Jan 3, 2026, we wanted to share how our associated Hivemind exposure performed in excess of market and industry for US-listed stocks:

• Jan 5 (post-capture) vs Jan 2 (pre-capture): +1.2%
• Jan 6 vs Jan 5: +0.3%
• Jan 6 vs Sep 25 (since becoming a top Hivemind trend): +5.2%

Exposure return evolution (market and industry regressed out):

Trend popularity evolution:

To learn more about Hivemind, or to schedule a demo, email us at hi@ forecastos.com

Suggested Reading From the ForecastOS Blog:

• Hivemind Factors Explain Incremental Out-of-Sample Variance
• How Stable Are Thematic Hivemind Exposures?
• ForecastOS Hivemind: Measuring Emergent Sources of Risk & Return Since 2020
• ForecastOS Hivemind: Understand Your Exposures

The answer, of course, depends on how our customizable exposure pipeline is tuned. For this article, we'll restrict the pipeline to annual data inputs only, and analyze changes year over year for an R3 proxy universe. We'll do so for 3 thematic exposures since 2023:

• AI and GenAI
• Inflation
• US Tariffs

Table of Contents:

1. Exposure Stability: AI and GenAI
2. Exposure Stability: Inflation
3. Exposure Stability: US Tariffs
4. Takeaways

AI and GenAI has been a remarkably stable thematic Hivemind exposure over the last 3 years. Not a single value from our select large cap universe below changed, and few exposures changed from negative to positive, or vice versa, year over year.

Select Large Cap Exposures - AI and GenAI:

Exposure Transition Matrices - AI and GenAI:

2023 -> 2024 (counts and row %):

2024 -> 2025 (counts and row %):

Exposure Return - AI and GenAI:

Inflation has been a moderately stable thematic Hivemind exposure over the last 3 years. Half of the companies in our select large cap universe had an inflation exposure change at some point since 2023. Across all companies, negative exposures flipped to positive ~10% of the time, and positive exposures flipped to negative ~3% of the time.

In our experience, vague, challenging to quantify company exposures (like inflation) have more variance over time.

Select Large Cap Exposures - Inflation:

Exposure Transition Matrices - Inflation:

2023 -> 2024 (counts and row %):

2024 -> 2025 (counts and row %):

Exposure Return - Inflation:

Tariff exposures were fairly stable over the last 3 years. Across all companies, negative exposures flipped to positive <4% of the time. Positive tariff exposures were both rare and unstable.

Select Large Cap Exposures - US Tariffs:

Exposure Transition Matrices - US Tariffs:

2023 -> 2024 (counts and row %):

Transition matrix 2024 -> 2025 (counts and row %):

Exposure Return - US Tariffs:

• Clearly defined exposures are fairly stable throughout time
• More ambiguous or hard-to-quantify exposures are less stable, but not necessarily less valuable
• Certain thematic exposures (like tariffs) show both stability and specific quirks
• Exposure drift is partially explained by changes in company operations and disclosures

To learn more about Hivemind, or to schedule a demo, email us at hi@ forecastos.com

Table of Contents

1. Why 2025 Has Been Challenging For Quantitative PMs
2. Creating Macro and Subindustry Hivemind Factors
3. How We Measured Out-of-Sample Explained Variance (R²)
4. Results: Hivemind Factors Explain Incremental Out-of-Sample Variance
5. Email Us: hi@forecastos.com

If you ran a large book through 2025’s volatility, you probably spent more time on the phone than in the market.

“Are you down? We’re off 3%.”

Most of those calls were really a fishing expedition for: "what exposure did I miss?" and "does everyone have it, or just me?" The uncomfortable truth is that, outside of a handful of slow-moving market, industry, and style boxes, few managers could answer that question with conviction.

The challenging 2025 market environment has been quantifiable; it has been measurably harder to isolate alpha. Out-of-sample variance explained by market, industry / subindustry, and common style factors dropped by roughly a third vs previous benchmarks. Even sophisticated hedge funds like Citadel and Millennium have had lackluster years to date.

What could be causing this?

We believe the challenging 2025 investment environment was caused in part by growth in emergent macro factors and nascent subindustries. We outline and test this hypothesis below.

We believe out-of-sample explained variance declined in 2025 (in part) due to an increase in A) emergent macro factors and B) nascent subindustries not captured well by traditional factors.

A) Emergent Macro Factors

A quick look at the ForecastOS Hivemind UI gives us some ideas what macro factor exposures might be to blame:

Given the above, and what we learned from our most recent article Measuring Emergent Sources of Risk & Return Since 2020, we'll use the following Hivemind stock-specific macro exposures to better capture volatility:

• AI / generative AI
• Defense and national security spending
• Green energy
• Tariffs
• Medicaid cuts
• Inflation

Select large cap Hivemind macro exposures:

Tariff exposure return (2025):

B) Identifying Nascent Subindustries

For nascent subindustries, we can create a Hivemind pipeline to describe each company in a custom, standardized way.

We can then feed this Hivemind standardized output into an unsupervised learning algorithm to systematically identify groups of similar companies. Doing so at our preferred level of specificity, we identify 292 nascent subindustries across several thousand US listed companies.

If a company does it, we've identified it, and our process is resilient to new / changing businesses going forward!

Select large cap Hivemind nascent subindustries:

Using our aforementioned 6 Hivemind macro factors and our 292 new Hivemind subindustries, we can measure incremental out-of-sample explained variance.

Benchmark: The Traditional Factors We Used

To perform our analysis, we used the following traditional factors: a market factor (1 for all companies; our intercept), industry (12) and subindustry (32) factors from a major data vendor (one-hot encoded), and style factors.

To calculate our QVM (quality, value, momentum) and size style factors, we used the below ForecastOS feature engineering config:

feature_normalization:
  winsorize:
    - 0.05
    - 0.95
    - - datetime
  standardize:
    - datetime

feature_adjustments_post_normalization:
  shift: 
    - 20 # Avoid lookahead
    - - id
      - datetime
    - id
  zero_fill: []

features:  
  # Quality: margin
  gross_income_margin_ltm:
    uuid: 49c61109-c029-4c4d-85c9-b1f33a72ea50
  ebit_margin:
    uuid: 47d7d1b0-46cc-4620-b0d3-35aa33fe490a
  net_income_margin_ltm:
    uuid: 7f1e058f-46b6-406f-81a1-d8a5b81371a2

# ...
# Email us for full ~200 line config

Note that we only use style_quality, style_value, style_momentum, and style_size in our regression; other factors listed above are simply inputs to the above composite QVM style factors.

Our Regression

Our regression was cross-sectional for the top 1000, 2000, and 3000 US stocks by market cap, fitting a separate model for each 20-trading-day return period. To control instability from correlated or thinly-supported factors, we applied ridge regularization. The shrinkage penalty tempered extreme coefficients, ensuring that each exposure competed for explanatory power without overfitting to noise.

Further, to prevent wild coefficients for the 292 emerging sub-industry factors, we capped their effective influence through both ridge shrinkage and scaled exposure weighting. In practice, this meant thin sub-industries were allowed to express signal - but only in proportion to the data depth supporting them, keeping the regression stable.

For each date, we split the universe into two disjoint sets: one for training (75%), where coefficients were estimated, and another for testing (25%), where regression performance (i.e. explained variance or R²) was measured. This intra-period separation helped assess generalization i.e. whether the fitted relationships held beyond the specific sample used for estimation.

As per above, Hivemind’s macro and subindustry factors incrementally improved out-of-sample explained variance meaningfully; they identified systematic risk that conventional factors missed.

Perhaps unsurprisingly given dynamic macro and market narratives as of late, the improvement was stronger in 2025 than in prior years - roughly double.

In today’s fast-moving investment landscape, where generative AI, inflation fluctuations, geopolitical upheavals, and global trade shifts increasingly dictate market behaviour, static risk models fall short. Only tools rooted in perceived causality, not outdated correlations, can help institutional investors quickly and accurately surface and manage dynamic exposures.

That’s exactly what ForecastOS Hivemind delivers. By converting unstructured data - podcasts, filings, and other live feeds - into clean, point‑in‑time factor signals, Hivemind equips you with real-time insights for emergent themes.

ForecastOS Hivemind isn’t just a tool - it’s your edge. In an era where the pace and complexity of market forces are escalating, Hivemind empowers you not just to respond, but to anticipate and act with conviction.

Let’s partner to help you discover, measure, and manage emergent sources of alpha and risk!

To learn more about Hivemind, or to schedule a demo, email us at hi@ forecastos.com

Table of Contents:

1. ForecastOS Hivemind
2. Emergent Themes; Top Sources of Risk & Return Since 2020
3. Closing Thoughts

Now, more than ever, the stock market is driven by emergent themes:

• The rise of generative AI
• Rapidly evolving inflation environments
• Wars, both trade and actual
• Rapidly shifting political agendas

A traditional, structural risk model can't tell you how exposed your portfolio is to these new-school sources of risk and return. Even sub-industry-neutral portfolios remain heavily exposed.

You, a smart, institutional investor, know to actively and intentionally manage your portfolio's exposures. The problem is that the risk and performance attribution tools you're using can't identify and manage these increasingly frequent, emergent themes that drive performance and volatility.

Hivemind, ForecastOS’s new exposure functionality, was built precisely for that blind spot.

For institutional investors, this means you can:

• Explore, pinpoint, and quantify novel risks and opportunities with precision
• Seamlessly incorporate emerging trend exposures into your risk monitoring, attribution, or alpha-signal pipelines
• Move faster than the markets; proactively identifying blind spots before they become fixations

Let's explore how Hivemind both surfaces and creates exposures for point-in-time emergent trends next!

Under the hood, Hivemind is a generative-AI engine that turns unstructured data firehoses (podcasts, filings, etc.) into clean, point-in-time factors. Instead of forcing your book into a static factor coffin, Hivemind lets you explore emergent point-in-time trends...

quantify trend movement...

turn emergent trends into exposures...

and even define new exposures for your equity universe...

Want to know which of your holdings are truly tariff-sensitive, or how exposed you are to rising inflation?

Describe the concept once and Hivemind scores every name in your universe, as frequently as every day. These bespoke exposures flow straight into your risk, performance-attribution, or signal pipelines in minutes, not months.

Using Hivemind, let's surface and examine the most prominent, market relevant trends since 2020.

We can use Hivemind's trend aggregation tool to rank market relevant trends since 2020...

... and it looks like the most prominent market-relevant trends since 2020 were:

1. COVID-19 pandemic
2. Conflict: Russia-Ukraine
3. U.S. elections and associated policy
4. Generative artificial intelligence
5. Inflation and interest rates
6. Conflict: Middle East
7. Cryptocurrency
8. Conflict: China-Taiwan
9. U.S. tariffs (from page 2)
10. Medicaid cuts (from page 2)

Let's explore each of them below, using Hivemind to measure discussion volume and associated exposure returns!

1. COVID-19 Pandemic

In early 2020, COVID-19 spread worldwide, leading the WHO to declare a global pandemic. The rollout of vaccines / mass vaccinations started in late 2020, marking a major turning point. By early 2023, restrictions were largely lifted, and the WHO declared the public health emergency over.

COVID-19 Pandemic Annotations

2. Conflict: Russia-Ukraine

Russia’s full-scale invasion of Ukraine in early 2022 sparked a devastating war that drew global attention and support for Ukraine. The conflict remains ongoing, marked by heavy casualties, economic disruption, and deepening geopolitical tensions.

Conflict: Russia-Ukraine Annotations

3. U.S. Elections & Associated Policy

The 2020 election and its aftermath highlighted deep political divisions, culminating in the Capitol attack and Trump’s second impeachment. Shortly thereafter, Republicans regained power in Congress during the 2022 midterms. In 2024, Trump won re-election and was inaugurated in January 2025.

Note: tariffs (trend #9) and Medicaid cuts (trend #10) are reviewed separately later in article as well.

U.S. Elections & Associated Policy Annotations

4. Generative Artificial Intelligence

Since 2020, Gen. AI has been the largest economically-driven investment theme. The public launch of ChatGPT in November 2022 made AI mainstream overnight, sparking widespread adoption. Other companies quickly followed suit. The technology has advanced rapidly alongside concerns from governments and other stakeholders regarding potential risks and societal impact.

Note: Google’s 2017 paper “Attention Is All You Need,” introduced the transformer architecture used by ChatGPT and other LLMs.

Generative Artificial Intelligence Annotations

5. Inflation & Interest Rates

The COVID-19 pandemic triggered aggressive fiscal and monetary policy, causing inflation to surge throughout 2021 and 2022. In early 2022, the Fed began hiking rates to combat inflation. Recently, as the US economy has begun to show signs of weakness, the Fed has signalled it's thinking about slashing rates.

Inflation & Interest Rates Annotations

6. Conflict: Middle East

Conflict in the Middle East escalated sharply after the Oct 7, 2023 Hamas attack on Israel. The conflict has since grown into a broader, multi-front confrontation involving Iran, the U.S., and various proxy forces.

Middle East Conflict Annotations

7. Cryptocurrency

Since 2020, cryptocurrency has surged in popularity. This growth has been accompanied by numerous scandals, such as high-profile rug pulls, the FTX collapse, and major exchange hacks. In response, governments worldwide have increasingly implemented regulations to enhance oversight, protect investors, and curb illicit activity.

Note: the above exposure return is (intuitively) correlated with BTC and S&P500 returns.

Cryptocurrency Annotations

8. Conflict: China-Taiwan

Over the past five years, Taiwan has faced mounting pressure from China through military drills and diplomatic isolation. The U.S. has shown consistent support through arms sales and high-profile visits, further straining cross-strait relations. These escalating tensions risk sparking conflict in a region vital to global trade, security, and (Taiwanese) semiconductor supply chains.

Note: a war has not broken out (yet) in this region. As such, the exposure return above is fairly flat (see y-axis), as expected.

China Taiwan Conflict Annotations

9. U.S. Tariffs

A key aspect of Donald Trump's political platform has been aggressive and widespread tariff use. These measures have targeted both allies and rivals, sparking retaliation and raising import costs to historic levels. A growing body of people doubt his resolve and follow through, leading to the rise of the acronym TACO (Trump Always Chickens Out).

U.S. Tariffs Annotations

10. Medicaid Cuts

Medicaid funding has changed over time, largely related to which political party is in power and (the end of) COVID-19 related initiatives.

Medicaid Cuts Annotations

In today’s fast-moving investment landscape, where generative AI, inflation fluctuations, geopolitical upheavals, and global trade shifts increasingly dictate market behaviour, static risk models fall short. Only tools rooted in perceived causality, not outdated correlations, can help institutional investors quickly and accurately surface and manage these dynamic exposures.

That’s exactly what ForecastOS Hivemind delivers. By converting unstructured data - podcasts, filings, and other live feeds - into clean, point‑in‑time factor signals, Hivemind equips you with real-time insights for emergent themes.

Want to see how tariff risk is evolving across your portfolio today? Or how exposed you are to sharp inflation shifts? Hivemind lets you generate and score exposures across your universe in minutes - not months - even for brand‑new trends.

For institutional investors, this means you can:

• Discover and quantify novel risks and opportunities with precision
• Seamlessly incorporate emerging trend exposures into your risk monitoring, attribution, or alpha pipelines
• Move faster than the market

ForecastOS Hivemind isn’t just a tool - it’s your edge. In an era where the pace and complexity of market forces are escalating, Hivemind empowers you not just to respond, but to anticipate and act with conviction.

Let’s partner to help you discover, measure, and manage emergent sources of alpha and risk!

To learn more about Hivemind, or to schedule a demo, email us at hi@ forecastos.com

“Are you down? We’re off 3%.”

Most of those calls were really a fishing expedition for: "what exposure did I miss?" and "does everyone have it, or just me?" The uncomfortable truth is that, outside of a handful of slow-moving style boxes, few managers can answer that question with conviction.

Traditional factor models were designed for a different era. MSCI’s style, industry, and macro factors do a respectable job explaining pre-2020-era performance drivers, but they struggle to keep pace with the meme-speed economics that now push risk on and off balance sheets overnight.

Hivemind: The Gen-AI Trend & Exposure Engine

Hivemind, ForecastOS’s new exposure functionality, was built precisely for that blind spot.

Under the hood, Hivemind is a generative-AI engine that turns unstructured data firehoses (podcasts, filings, etc.) into clean, point-in-time factors.

Instead of forcing your book into a static factor coffin, Hivemind lets you explore emergent trends...

quantify trend movement...

and define what exposure matters...

Want to know which holdings are truly tariff-sensitive, or how exposed you are to rising inflation?

Describe the concept once and Hivemind scores every name in your universe, as frequently as every day. Those bespoke exposures flow straight into your risk, performance-attribution, or signal pipelines in minutes, not months.

30 Second Case Study: Inflation Shock

When a U.S. hedge fund needed to attribute performance and dampen volatility from shifting rates and inflation rhetoric, they couldn't find a dataset that captured which equities in their ~3,000 company universe were actually at risk.

Hivemind made this inflation attribution for their universe in a single pass: ingest point-in-time data, then output a ready-to-model “inflation exposure” factor that re-scores companies whenever the conversation changes.

Hivemind: From “What Happened?” to “What’s Next?”

Because Hivemind is point-in-time, you can replay exposures exactly as they were understood before the print.

Tie that to ForecastOS Canary and you get live alerts: when a theme you care about bubbles up, when new market-relevant risk factors start to form, or when an assumption your strategy leans on starts to wobble.

Instead of discovering the driver after the drawdown, you see it forming, quantifiable, in advance.

A quick list of noteworthy trends Hivemind identified from point-in-time data in advance of market movement:

• COVID-19
• ZIRP era inflation
• Middle-eastern conflict and Iran strike
• Russia-Ukraine conflict
• AI / Generative AI (as an emerging trend and investment theme)
• Medicaid and Medicare cuts

Hivemind: Why It Matters Now

Macro is fragmenting, information velocity is compounding, and “risk-off” is no longer a single switch.

In that landscape, edge comes from knowing which narratives are breaking out, and when - before the spreadsheet says so. Hivemind lets your team retire the phone-tag post-mortems and deploy capital with a live, quantifiable map of what’s really moving your portfolio.

We built Hivemind to make every line item as transparent as your best-researched theme - so you can stop guessing at exposures and start engineering them!

Creating a Macro Investment Strategy Overlay with Hivemind

Speaking of engineering exposures, check out our below demo to see how you can use Hivemind to create an S&P 500 macro strategy overlay for your portfolio with:

• 2.5% excess return annualized
• 1.0x information ratio
• 3.4x one-way turnover
• ~50% target active weight vs benchmark

For trial access: hi@forecastos.com

A Closing Thought

Timing (macro / thematic) trends in the market is largely considered to be a fool's errand... but we believe it's a complicated (but possible) data and analytics problem.

The problem, at its core, consists of:

• Identifying trends (via quantifying what everyone thinks at each point-in-time), and
• Generating associated company exposures (i.e. quantifying which companies are impacted positively and negatively).

Hivemind is our solution to this data and analytics problem. We're excited to see what you do with it!

In this blog post, we use Hivemind to create a time series factor based on changes in aggregate sentiment about whether Elon Musk is a genius. We calculate how this factor correlates with Tesla’s forward stock returns and use it as a signal in a single-stock long/short trading strategy.

Agenda

1. What is ForecastOS Hivemind?
2. Hivemind Poll: "Is Elon Musk a Genius?"
3. Elon - TSLA Correlation
4. "Is Elon Musk a Genius?" Trading Strategy
5. Build Your Own Predictive Signals With Hivemind
6. Full Code

1. What is ForecastOS Hivemind?

ForecastOS Hivemind is an AI-powered platform that continuously ingests data from top media and financial sources. It allows you to create time series factors from unstructured data using generative AI.

If you can identify a topic that influences a stock’s value, Hivemind can track consensus over time and convert it into a signal with alpha potential.

Hivemind can also identify emerging trends and generate company-level risk exposures. For example, it recently flagged the Iran strike days before it occurred and identified the stocks likely to be positively or negatively impacted.

2. Hivemind Poll: "Is Elon Musk a Genius?"

73,315 mentions of Elon Musk were evaluated to assess whether speakers agreed with the statement, "Elon Musk is a genius." Responses generated by AI were scored as 1 for yes, 0 for neutral or no opinion, and -1 for no. Rolling averages over 90 and 365 days were then calculated to produce the graph shown below.

3. Elon - TSLA Correlation

Using 90-day rolling average “Elon Musk is a genius” sentiment and Tesla stock price data from ForecastOS FeatureHub, we examined if sentiment shifts were correlated with forward stock returns.

The 1-day sentiment change isn’t statistically significant, likely due to higher noise / volatility at that time horizon. The 30-day change, however, shows strong statistical significance!

4. "Is Elon Musk a Genius?" Trading Strategy

Given the positive correlation between changes in Elon Musk genius sentiment and Tesla stock returns, we tested a simple trading strategy using sentiment changes as a signal.

The strategy goes long when sentiment rises and short when it falls. Below is the performance graph for the strategy.

To show our performance is not simply due to the percentage of positive and negative positions, especially since Tesla’s price rose ~30x over the backtest period, we benchmarked against a randomized strategy. This strategy takes random positions with the same positive and negative ratio as our sentiment factor. Below is the performance graph for the randomized strategy.

The strategy that uses our sentiment factor outperforms the randomized version by roughly 3x, confirming the predictive value of our sentiment signal.

While we wouldn’t advocate running this strategy on its own, especially due to its high volatility, turnover, and associated trading costs, it’s an example of a great addition / input to a wider fundamental or quantitative investment strategy!

5. Build Your Own Predictive Signals With Hivemind

This is just one example of how Hivemind can be used to easily create forward-predictive signals that generate alpha. With Hivemind, any topic can become a time series factor, providing you with new, customizable risk and alpha inputs.

If you want to build your own signals, such as risk exposures to tariffs or recession likelihood sentiment, Hivemind makes it easy. Click below to start your free trial today!

6. Full Code

import pandas as pd
import os, forecastos as fos
import numpy as np
from scipy.stats import pearsonr

# -------------------------------------------------------
# 1. Get Elon sentiment data
# -------------------------------------------------------

# Poll results generated by Hivemind
df_elon_sentiment = pd.read_csv('./elon_poll.csv') 
df_elon_sentiment['episode_date'] = pd.to_datetime(df_elon_sentiment['episode_date'])

# Create a full date range from min to max date
full_range = pd.date_range(
  start=df_elon_sentiment['episode_date'].min(), 
  end=df_elon_sentiment['episode_date'].max(), 
  freq='D'
)

# Reindex the DataFrame with the full date range
df_full = df_elon_sentiment.set_index('episode_date').reindex(full_range)

# Rename the index to 'episode_date'
df_full.index.name = 'episode_date'

# Forward fill the rolling_avg column
df_full['rolling_avg'] = df_full['rolling_avg'].ffill()

# Reset index
df_full = df_full.reset_index()

# Shift by 1 day to avoid lookahead 
# --> (sentiment comes in over entire day for datetime, using open to open returns)
df_full['rolling_avg'] = df_full['rolling_avg'].shift(1)

# Update sentiment data to include all dates
df_elon_sentiment = df_full

# Calculate changes in sentiment over different time windows
df_elon_sentiment["rolling_avg_prev_1d_diff"] = df_elon_sentiment["rolling_avg"] - df_elon_sentiment["rolling_avg"].shift(1)
df_elon_sentiment["rolling_avg_prev_2d_diff"] = df_elon_sentiment["rolling_avg"] - df_elon_sentiment["rolling_avg"].shift(2)
df_elon_sentiment["rolling_avg_prev_3d_diff"] = df_elon_sentiment["rolling_avg"] - df_elon_sentiment["rolling_avg"].shift(3)
df_elon_sentiment["rolling_avg_prev_4d_diff"] = df_elon_sentiment["rolling_avg"] - df_elon_sentiment["rolling_avg"].shift(4)
df_elon_sentiment["rolling_avg_prev_5d_diff"] = df_elon_sentiment["rolling_avg"] - df_elon_sentiment["rolling_avg"].shift(5)
df_elon_sentiment["rolling_avg_prev_10d_diff"] = df_elon_sentiment["rolling_avg"] - df_elon_sentiment["rolling_avg"].shift(10)
df_elon_sentiment["rolling_avg_prev_30d_diff"] = df_elon_sentiment["rolling_avg"] - df_elon_sentiment["rolling_avg"].shift(30)
df_elon_sentiment["rolling_avg_prev_90d_diff"] = df_elon_sentiment["rolling_avg"] - df_elon_sentiment["rolling_avg"].shift(90)

df_elon_sentiment = df_elon_sentiment.rename(columns={"episode_date": "datetime"})
df_elon_sentiment['datetime'] = pd.to_datetime(df_elon_sentiment['datetime'])
df_elon_sentiment["ticker"] = 'TSLA' + '-US'

# -------------------------------------------------------
# 2. Join ticker mapping and returns data from FeatureHub
# -------------------------------------------------------

fos.api_key = os.environ.get("FORECASTOS_API_KEY_PROD")

# Ticker mapping data
df_id_mapping = fos.Feature.get(
  "79f61521-381e-4606-8020-7a9bc3130260"
).get_df().rename(columns={"value": "ticker"})

df_elon_sentiment = df_elon_sentiment.merge(df_id_mapping, on='ticker', how='left')
df_elon_sentiment['ticker'] = df_elon_sentiment['ticker'].str[:-3]

# Merge 5d returns data into df
df_prev_5d_return = fos.Feature.get(
  "8700c7c9-4d14-400b-a27b-93605d3eadbf"
).get_df()

df = df_elon_sentiment.merge(df_prev_5d_return, on=['id', 'datetime'], how='left').rename(columns={
    "value": "prev_5d_return"
})

# Merge 1d returns data into df
df_prev_1d_return = fos.Feature.get(
  "ea4d2557-7f8f-476b-b4d3-55917a941bb5"
).get_df()

df = df.merge(df_prev_1d_return, on=['id', 'datetime'], how='left').rename(columns={
    "value": "prev_1d_return"
})

df['prev_5d_return'] = df['prev_5d_return'].fillna(0)
df['fwd_5d_return'] = df['prev_5d_return'].shift(-5)
df['prev_1d_return'] = df['prev_1d_return'].fillna(0)
df['fwd_1d_return'] = df['prev_1d_return'].shift(-1)

df = df.dropna(subset=['rolling_avg_prev_2d_diff', 'fwd_5d_return', 'fwd_1d_return'])

# -------------------------------------------------------
# 3. Correlations
# -------------------------------------------------------

# --- 1 day sentiment diff correlation to returns --- #

# Drop rows with NaNs in either column
clean_df = df[['rolling_avg_prev_1d_diff', 'fwd_5d_return']].dropna()

# Compute correlation coefficient and p-value
r, p_value = pearsonr(clean_df['rolling_avg_prev_1d_diff'], clean_df['fwd_5d_return'])

print(f"Correlation coefficient (r): {r:.4f}")
print(f"P-value: {p_value:.4e}")

# --- 30 day sentiment diff correlation to returns --- #

# Drop rows with NaNs in either column
clean_df = df[['rolling_avg_prev_30d_diff', 'fwd_5d_return']].dropna()

# Compute correlation coefficient and p-value
r, p_value = pearsonr(clean_df['rolling_avg_prev_30d_diff'], clean_df['fwd_5d_return'])

print(f"Correlation coefficient (r): {r:.4f}")
print(f"P-value: {p_value:.4e}")

# -------------------------------------------------------
# 4. Simple strategy backtest
# -------------------------------------------------------

df['position'] = np.where(df['rolling_avg_prev_1d_diff'] >= 0, 0.2, -0.2)
df['position_rolling_sum_5'] = df['position'].rolling(window=5).sum()
df['strategy_return_fwd_1d'] = df["position_rolling_sum_5"] * df["fwd_1d_return"]
df['strategy_aum'] = (df['strategy_return_fwd_1d'] + 1).cumprod()

df.set_index('datetime')['strategy_aum'].plot(title='"Elon Musk Is a Genius" Strategy Performance', xlabel='Date', ylabel='Strategy AUM', grid=True)

# --------------------------------------------------------------------
# 5. Backtest with same amount negative/positive positions, but random
# --------------------------------------------------------------------

percent_negative = (df["fwd_1d_return"] < 0).mean() * 100
percent_postive = (df["fwd_1d_return"] >= 0).mean() * 100

position_negative_pos = (df["position_rolling_sum_5"] < 0).mean() * 100
position_positive_pos = (df["position_rolling_sum_5"] >= 0).mean() * 100

n = len(df)
n_neg = int(round(position_negative_pos / 100 * n))
n_pos = n - n_neg  # ensures sum to n

# Create the values
values = np.array([-0.2] * n_neg + [0.2] * n_pos)

# Shuffle the values to randomize their positions
np.random.seed(42)
np.random.shuffle(values)

# Add as a new column
df['position_random'] = values
df['strategy_return_fwd_1d'] = df["position_random"] * df["fwd_1d_return"]
df['strategy_aum'] = (df['strategy_return_fwd_1d'] + 1).cumprod()

df.set_index('datetime')['strategy_aum'].plot(title='Randomized Strategy Performance (Equal Long/Short Weights)', xlabel='Date', ylabel='Strategy AUM', grid=True)

When a U.S. hedge fund asked us to help with research to dampen volatility from shifting tariff rhetoric in the Trump era, the first road-block was obvious: no comprehensive dataset captured which of the most active ~3,000 U.S. equities were truly exposed to tariffs, let alone gave the knobs to dial those exposures up or down.

Fortunately, we already had the tool for the job: ForecastOS Hivemind.

What is ForecastOS Hivemind?

Hivemind is ForecastOS’ generative-AI engine for turning unstructured time series data into clean, point-in-time factors.

Under the hood, it ingests everything from SEC filings (or any text / financial information) to top podcasts (or any video / audio). It then indexes information by meaning, context, datetime, and associated company (where relevant).

Using Hivemind, you can define, customize, and score any concept throughout time, using any underlying dataset, whether it's tariff or AI exposures, CEO or research analyst optimism, etc.

Because factor recipes are just a few clicks or API calls away, quants can spin up bespoke signals in minutes instead of months!

Why Did We Build ForecastOS Hivemind?

Traditional factor libraries miss the idiosyncratic, fast-moving themes that drive returns and volatility today. Hivemind closes that gap by letting researchers easily generate, test, and deploy fresh, completely customizeable factors on demand, using both public and proprietary data.

Try ForecastOS Hivemind Out!

Whether you want to create tariff, AI / chip, or geopolitical-hot-spot exposures, just define:

• what you want scored,
• how you want it scored,
• your result schema (i.e. what format you want the factor data to be in), and
• which ForecastOS datasets you want to use,

and let ForecastOS Hivemind do the rest. For thematic exposures, sentiment, popularity, or any time series factor!

Demo and sandbox key available via trialaccess @ forecastos.com. Let's make the GPUs go brrr!

By the end of this post, we'll have a model that predicts 10-trading-day forward changes to VIX during our test period (last ~2 years) with:

• 63.8% hit rate (directional accuracy),
• 13.2% R² score (explained variance),
• 0.37 Pearson correlation, and
• 0.46 Spearman correlation.

Agenda

1. Introduction: VIX and Hivemind
2. How to: create custom Hivemind time-series factors to predict VIX (or anything)
3. Analysis: custom Hivemind factor correlations with forward VIX
4. Analysis: previous VIX momentum correlations with forward VIX
5. Analysis: composite Hivemind factor correlations with forward VIX
6. Analysis: ML for VIX forecasts
7. Suggested improvements
8. Why we built Hivemind
9. Closing / contact info

1. Introduction: VIX and Hivemind

1.1 VIX (Volatility Index)

The VIX (Volatility Index or fear index colloquially) measures expected market volatility over the next 30 days using prices of a specific group of S&P 500 call and put options. Forecasting the VIX is crucial for building robust risk models, trading financial derivatives, or when using software like InvestOS, where volatility estimates directly affect portfolio construction and exposure scaling.

1.2 ForecastOS Hivemind; Creating Factors

ForecastOS Hivemind is a proprietary tool for creating popularity and sentiment-based time-series factors from point-in-time audio/video and text/written content. Currently, the tool predominantly uses top US podcasts as audio input and SEC filings as text input.

2. How To: Create Custom Hivemind Time-Series Factors to Predict VIX (or Anything)

Today we are forecasting changes to future VIX levels, so let's create custom 90-day Hivemind popularity factors for:

• Anxiety / worry levels ("anxiety")
• Job security / economic concerns ("unemployment")
• Bearish market outlook ("market crash")
• Geopolitical conflict ("global conflict")

2.1 Creating Hivemind Factors

We can use the following code to create our custom Hivemind time-series factors:

import requests
import os

FACTORS_TO_TEST = [
    "anxiety",
    "unemployment", 
    "market crash",
    "global conflict",
] 

# URL and headers
url = "https://app.forecastos.com/api/v1/trends/custom"
headers = {
    "Authorization": f"Bearer {os.getenv('HIVEMIND_API_KEY')}",
    "Content-Type": "application/json"
}

results = {}
for hivemind_factor in FACTORS_TO_TEST:
    data = {
        "trend": {
            "text": hivemind_factor,
            "sensitivity": "medium", # For tuning meaning-based similarity of matches
        }
    }

    response = requests.post(url, json=data, headers=headers)
    results[hivemind_factor] = response.json()

2.2 Viewing a Hivemind Time-Series Factor

Using the Hivemind UI (or a plotting library like matplotlib) we can view the popularity evolution and associated mentions for any custom factor, like global conflict:

Note the spikes in 90 day popularity preceding and during the onset of the Russia-Ukraine war, US tariffs, etc.

3. Analysis: Custom Hivemind Factor Correlations with Forward VIX

Let's analyze the correlations between our Hivemind-generated factors and forward 10-day changes in VIX from 2016 to present. These raw correlations and associated p-values should help us identify and validate that our factors are leading indicators of volatility spikes.

While we will use 10, 15, 20, 25, and 30 day (backwards-looking) factor growth for making forward VIX growth predictions, we only show 20 day (backwards-looking) factor growth below to save space.

Anxiety levels are not strongly rank-order (Spearman) correlated on their own with forward changes to VIX. However, given this factor is intuitive and has a high Pearson correlation, we are going to keep it; it will help our non-linear modelling efforts later in this article.

Let's also quickly confirm our delta Hivemind factors aren't too (Pearson) correlated with each other:

They are not, and as such should all be additive to our forecasts. Excellent!

4. Analysis: Previous VIX Momentum Correlations with Forward VIX

Often, previous changes in the target variable itself (i.e. momentum factors) are correlated with forward changes. Given the VIX is mean-reverting, sign-flipped previous changes should be correlated. Let's explore that below.

They are, and as such should all be additive to our forecasts. Great!

5. Analysis: Composite Hivemind Factor Correlations with Forward VIX

Next, let's create composite factors using weighted combinations of our delta factors over a mix of growth horizons. They should provide smoother and more stable signals.

6. Analysis: ML For VIX Forecasts

Using all of the factors we've created thus far, let's use XGBoost to forecast our target variable: 10-day (forward) changes in VIX.

We'll use the oldest 75% of dates as our training set and newest 25% of dates (less a 10 day gap to avoid lookahead) as our test set.

6.1 Train XGBoost Model to Predict Forward VIX Changes

import xgboost as xgb

import numpy as np
np.random.seed(42)

TARGET = 'fwd_10d_growth_VIX'

# Copy the original time-series factor DataFrame for ML use
ml_df = merged_df.copy()

# Drop rows with NaNs
ml_df = ml_df.dropna()

# Determine split index (75% train, 25% test)
split_idx = int(len(ml_df) * 0.75)

# Train / test split (df already sorted oldest -> newest)
train = ml_df.iloc[:split_idx]
test = ml_df.iloc[split_idx + 10:]

X_train = train.drop(columns=[col for col in ml_df.columns if "fwd" in col])
y_train = train[TARGET]
X_test = test.drop(columns=[col for col in ml_df.columns if "fwd" in col])
y_test = test[TARGET]

# Train model
model = xgb.XGBRegressor(
    n_estimators=1000, 
    max_depth=4, 
    learning_rate=0.003,
    subsample=0.6,
    colsample_bytree=0.8,
    colsample_bylevel=0.8
)
model.fit(X_train, y_train)

# Predict
y_pred = model.predict(X_test)

6.2 Visualize Predictions vs Actuals

It's often easier to see performance, so let's visualize how we did below.

Evolution:

Scatterplot:

Looks like we've done fairly well! Let's quantify our performance next.

6.3 Quantify Performance

Running the below code, we get:

• 63.8% hit rate (directional accuracy)
• 13.2% R² score (explained variance)
• 0.37 Pearson correlation
• 0.46 Spearman correlation

import numpy as np
from scipy.stats import spearmanr, pearsonr
from sklearn.metrics import r2_score

y_hat = y_pred
y = y_test.values

# Hit rate (directional accuracy)
hit_rate = np.mean(np.sign(y_hat) == np.sign(y))
print(f"Hit rate: {hit_rate:.2%}")

# Explained variance (R²)
r2 = r2_score(y, y_hat)
print(f"R² score (change in VIX): {r2:.4f}")

# Correlations
pearson_corr, pearson_pval = pearsonr(y_hat, y)
spearman_corr, spearman_pval = spearmanr(y_hat, y)
print(f"Pearson correlation: {pearson_corr:.4f} (p-value: {pearson_pval:.4g})")
print(f"Spearman correlation: {spearman_corr:.4f} (p-value: {spearman_pval:.4g})")

7. Suggested Improvements

• Add more features / factors: market-based, macro / economic indicators, VIX term structure, alternative data, Hivemind sentiment factors, etc.
• Train ML model on a rolling-window basis with walk-forward validation
• Ensemble predictions across models trained at different time horizons
• Grid-search hyperparameters
• Try other ML models
• Run SHAP to better understand forecast drivers and remove features that aren't sufficiently additive / helpful to reduce noise

8. Why We Built Hivemind

Imagine knowing what consensus views were, and how they were evolving, about anything, throughout time.

Creating accurate forecasts for anything would be easy with perfect access to features / factors representing aggregate popularity and sentiment for anything. The market, after all, is just aggregate sentiment.

However, this aforementioned humankind popularity and sentiment factor forge doesn't exist, and so creating accurate forecasts is hard.

We (very naively) thought we could create a tool using new developments in AI that allowed anyone to create well-founded time-series factors for anything in one line of code.

We were wrong. For several months and several disappointing Hivemind iterations.

But recently, after lots of improvements to our AI, software, and data architecture, Hivemind has started to work. We've built a humankind popularity and sentiment factor forge for any time-series factor you can imagine!

Right now, it's predominantly based on discussion from top US podcasts, but we will continue to add new (quality, non-AI generated) media sources to strengthen the Hivemind.

While our popularity factor forge is live today (and was used to create forecasts in this article), our sentiment factor forge is currently being finalized; expect to see it in the next couple of weeks!

We built Hivemind to allow anyone to create any popularity or sentiment time-series factor in one line of code; we're excited to see what you do with it!

9. Closing / Contact Info

Using ForecastOS Hivemind, we easily created custom factors and an associated model that predicts 10-trading-day forward changes to VIX during our test period (last ~2 years) with:

• 63.8% hit rate (directional accuracy),
• 13.2% R² score (explained variance),
• 0.37 Pearson correlation, and
• 0.46 Spearman correlation.

If you’re interested in using Hivemind for your own research / to create your own custom factors, drop us a line: trialaccess @ forecastos.com.

Notebook and code used for this article available to ForecastOS clients @ app.forecastos.com/research

Five noteworthy Hivemind trends from the past week:

1. Tariffs

Discussion focused on U.S.–China trade tensions, with concerns raised over economic uncertainty and rising consumer prices.

2. Trump / MAGA

Mentions of Donald Trump revolved around his involvement in Ukraine–Russia ceasefire talks, trade tensions with China, and public sentiment towards his leadership.

3. Artificial Intelligence / AI

Google drew attention in AI discussions this week over its AI search, recent earnings report, the potential forced sale of Google Chrome, and details from its antitrust trial about a deal with Samsung to preinstall the Gemini AI model.

4. Pope Francis

Discussion around the papacy spiked this week following the passing of Pope Francis, with conversations focusing on his legacy and anticipation for the upcoming conclave.

5. Stock Market Volatility

Discussion focused on market volatility and instability due to U.S. tariff policies.

Download the latest weekly research summary here:

If you’d like full access to Hivemind’s tools, trends, UI, and company-level insights, email us for a trial account (trialaccess @ forecastos.com).

--

Five noteworthy Hivemind trends from the past week:

1. Tariffs. Tariffs were the most prominent talking point this week, with particular focus on the escalating trade tensions between the U.S. and China.

2. NVIDIA / GPUs. Commentators responded to the newly announced export restrictions on NVIDIA’s GPU sales to China, discussing their potential impact on the company.

3. Deportation. Discussions on deportation revolved around the specific case of Kilmar Ábrego García, with commentators offering their views on both the case and current U.S. immigration policies.

4. Iran / Nuclear. Iran and its nuclear program received increased attention this week as the U.S. and Iran continued their nuclear deal negotiations.

5. Jerome Powell Call for Termination. Mentions of Jerome Powell centred around his comments on tariffs and Trump’s call for his termination.

➡️ Download the latest weekly research summary here:

--

If you’d like full access to Hivemind’s tools, trends, UI, and company-level insights, email us for a trial account (trialaccess @ forecastos.com).

Cheers!

--

Six noteworthy Hivemind trends from the past week:

1. Tariffs. The most discussed topic was tariffs, particularly Trump’s global tariff policy and his announcement of a 90-day pause on reciprocal tariffs.

2. Hamas, Israel, Gaza, and Palestine. The Israel-Gaza war remained a key topic this week, with conversations centering on Israel’s military expansion in Gaza, and on-going protests.

3. Trump. Mentions of Donald Trump focused on the impact of his policies, (specifically tariffs) on the U.S. economy, along with differing opinions about him.

4. Stock market. The stock market’s volatility in response to Trump’s tariffs was a major talking point, with commentators expressing concern over ongoing uncertainty and fears of long-term economic instability.

5. Panama Canal. Discussions about the Panama Canal’s importance to U.S. global policy and security have intensified this week as tensions rise due to China’s increasing interest in the region.

6. Nintendo. U.S. preorders for Nintendo’s newly announced Switch 2 were delayed due to Trump’s global tariffs, sparking increased conversations about Nintendo itself, the console’s availability, and potential price increases.

➡️ Access the latest edition here:

Stay tuned every week — and if you’d like full access to Hivemind’s tools, trends, and company-level insights, reach out for a trial or client account.

— The ForecastOS Research Team

Hivemind leverages advanced AI to process content from top media sources, identifying and analyzing tens of thousands of emerging trends. From geopolitical developments and economic shifts to cultural flashpoints and tech innovations, this dataset surfaces the stories shaping markets, policy, and sentiment.

While Hivemind is typically available only to ForecastOS clients, we’re making this weekly research open to all readers for the next several weeks. We believe this is a powerful moment to showcase the platform’s capabilities and share actionable intelligence more broadly.

The first edition covers trends from March 30 to April 6, including dramatic spikes in discussions on tariffs, the Israel-Gaza conflict, AI innovation, and Elon Musk’s continued political and economic influence.

We invite you to explore these weekly updates, see the signals before they become headlines, and consider what they might mean for markets, investment themes, and strategic decision-making.

➡️ Access the latest edition here:

Stay tuned every week — and if you’d like full access to Hivemind’s tools, trends, and company-level insights, reach out for a trial or client account.

— The ForecastOS Research Team